AI Reading for Tuesday July 22

OpenAI says ChatGPT users send over 2.5 billion prompts every day - The Verge

OpenAI’s new 'CEO for applications' strikes hyper-optimistic note - OpenAI

Incoming OpenAI Executive Fidji Simo Seeks Broader Market for AI Tools - Bloomberg

SoftBank and OpenAI’s $500 Billion AI Project Struggles to Get Off Ground - The Wall Street Journal

Stargate advances with 4.5 GW partnership with Oracle - OpenAI

GPT-5 hints rounded up, including a 'openai/gpt-5-reasoning-alpha-2025-07-13' config seen in the wild in a repo - BGR

Claude is increasingly dominant in coding. Was Anthropic cut-off the reason Windsurf's OpenAI deal imploded? - Business Insider

“Unfortunately, I think ‘No bad person should ever benefit from our success’ is a pretty difficult principle to run a business on,” wrote Anthropic CEO Dario Amodei in a note to staff saying they would take Middle East money. - WIRED

How AI agents can generate $450 billion by 2028 - and what stands in the way - ZDNET

Google says its system also earned gold in IMO, in competition, and with grading by competiton staff. - The New York Times

There has to be a catch and some level of hype here.

LLMs are poets, not quants, it’s a little like the English and Comp Lit folks taking over the math department. I am skeptical these new capabilities just emerged from scaling, without tools or some form of training set contamination, overfitting to the test.

Also, given hallucinations, if you can’t check it, you can’t use it. Why is this OpenAI guy so proud their model knows when it’s wrong, how could it do math without knowing that? Why no publicly available model or paper? And clearly OpenAI rushed a release to steal Google’s thunder.

Time will tell but I’ve done some math and done some LLM stuff and the two are just about orthogonal. I smell BS.

It reminds me of when 2 years Google said they beat like 85% of competitive coders and weren’t close. Although it’s now getting there with Claude and the OpenAI v. Psyho contest. And it is interesting that after all the dick-measuring between Google and OpenAI, Claude has been crushing them at coding.

Most of Meta's mega-hires were born in China, none in the US - The Economic Times

Is $100 million for an AI genius a bargain? - Business Insider

Trump’s immigration policies threaten American AI leadership. - Brookings

An anti-Sputnik moment, the dumbest people who can’t stand anyone who read a book that doesn’t contain comics, throwing away the US’s edge over nonsense.

China hawks: The Nvidia Chip Deal Trades Away the United States’ AI Advantage - Foreign Policy

Researchers tested LLMs using the Prisoner’s Dilemma game and found Google’s Gemini to be “strategically ruthless” and “strict and punitive,” while OpenAI’s ChatGPT was “catastrophically” cooperative and hopeful. - 404 Media

New Google experimental AI tool helps people explore the context and origin of images seen online, helping to verify them. - Google DeepMind

Google Seeks Licensing Talks With News Groups, Following AI Rivals - Bloomberg

Widely reported 15% drop in search referrals as AI use and AI summaries rise - Dataconomy

Chinese startup Manus launches dataviz creation feature, challenging ChatGPT code runner and agent features - VentureBeat

AMD and Stability AI released Stable Diffusion 3.0 Medium optimized for XDNA 2 NPUs, enabling local generation of 4MP images without GPU. - Tom's Hardware

I wonder if NPUs are a bust and APUs / IPUs integrating GPUs are the future.

Nvidia is making a new B30 model for China to comply with export restrictions - Tom's Hardware

Spotify Briefly Published AI-Generated Songs Atrributed To Dead Artists Without Permission - 404 Media

PayPal Adds Scam Detection to Friends and Family Payments - PYMNTS.com

AI tools in cold-chain warehouses cut spoilage via computer vision, optimized placement and and digital twins. - Business Insider

Last-minute addition of xAI to Pentagon contract raises concerns over safety, lack of vetted APIs and tooling, political meddling. - NBC News

Cuts off everyone else’s meds and research so millions gonna die, but never met government assistance for his own ventures he didn’t love and fight tooth and nail for.

XAI asked workers to record their facial expressions to train Grok — and they weren't happy - Business Insider

TFW you raise capital at $200b valuation on < $200m in revenue run rate - TechCrunch

Cursor AI safeguards easily bypassed in YOLO mode - The Register

AWS slaps usage caps on Kiro as AI editor preview proves too popular for its own good - The Register

NIH will cap researchers at six grant applications yearly after AI-generated proposals overwhelmed reviewers. - 404 Media

Meet the deepfake fraudster who applied to work in a deepfake detection group - ComputerWeekly.com

AI analysis of wastewater samples detects emerging viral variants earlier than clinical tests, aiding proactive public health responses. - News-Medical

Using AI to help design new physics experiments - Quanta Magazine

If you had an LLM in 2010, you could possibly have traded on news sentiment successfully. - LSE Eprints

Or you could train on Buffett’s writing corpus - Reddit

Douglas Rushkoff wants us to use AI to ask better questions - Fast Company

Anxious parents face tough choices on AI - Phys.org

The Hater's Guide To The AI Bubble. - Ed Zitron's Where's Your Ed At

It's good to be skeptical. There is an insane amount of hype that does everyone a disservice. At some point skepticism about stuff you can see with your own two eyes can do as much disservice as the hype.

Follow the latest AI headlines via SkynetAndChill.com on Bluesky